WO2023173543A1 - Data classification model training method and apparatus, classification method and apparatus, device, and medium - Google Patents

Abstract

The present application relates to a data classification model training method and apparatus, a classification method and apparatus, a device, and a medium. The training method comprises: dividing a plurality of data samples into a minority class sample set and a majority class sample set; undersampling from the majority class sample set to obtain an undersampling set; performing first iterative training on a classification model on the basis of a training set composed of the minority class sample set and the undersampling set to obtain a classification model meeting a first preset condition; if the model does not meet a second preset condition, performing oversampling on the minority class sample set on the basis of the model, and adding obtained samples into the training set; and performing second iterative training on the model on the basis of the updated training set to obtain a data classification model meeting the second preset condition. According to the training method in the present application, data obtained by means of undersampling and data obtained by means of oversampling are used for training the classification model, data used for training the classification model has good balance, a good training effect is achieved, and the classification accuracy of the trained classification model is high.

Description

Training methods, classification methods, devices, equipment and media for data classification models

priority statement

This application requires the priority of the Chinese patent application submitted to the China Patent Office on March 14, 2022, with the application number 202210248165.5, and the invention name is "Training method, classification method, device, equipment and medium for data classification model", all of which The contents are incorporated into this application by reference.

Technical field

This application relates to the field of artificial intelligence, and in particular to training methods of data classification models, data classification methods, devices, computer equipment and storage media.

Background technique

Data classification problem is one of the most common problems in the field of machine learning. Existing commonly used classification models include logistic regression algorithm model, k nearest neighbor algorithm model, decision tree algorithm model, support vector machine algorithm model, etc. As machine learning algorithms are used in more and more application scenarios, some problems have arisen in the application of classification models. Among them, due to the poor training effect of the classification model on imbalanced data, the classification accuracy of the trained classification model has been reduced. Not high, the imbalance of data distribution has a particularly significant impact on the classification effect. It is very difficult to obtain balanced data in some specific application scenarios. For example, in a telephone customer service scenario, there are very few complaint calls and many consultation calls. The number of calls between the two types differs by a hundred or even a thousand times, which brings great difficulties to training the customer complaint classification model. The inventor I realized that in the existing technology, historical data are directly used to train classification models. Since the historical data used for training is not processed in any way, the training effect is poor. The trained classification model will misidentify most of the complaint calls as consultation types. Phone,classification accuracy is low. Therefore, how to overcome the problems of poor training results and low classification accuracy of the trained classification model due to the imbalance of training data when training a classification model are currently technical issues to be solved.

Contents of the invention

Based on this, it is necessary to provide a training method for a data classification model, a data classification method, and a training method for a data classification model to address the problems of poor training results due to imbalance of historical data and low classification accuracy of the trained classification model when training the classification model. Devices, computer equipment and storage media.

A method for training a data classification model, including:

Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set;

Obtain an undersampled set from undersampling the majority class sample set;

Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition;

Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set;

Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition.

A data classification method that includes:

Get the data to be classified;

The steps of the above-mentioned data classification model training method; and,

The data to be classified is classified using the data classification model that satisfies the second preset condition.

A training device for data classification models, including:

A division module used to divide multiple pre-obtained historical data samples into a minority class sample set and a majority class sample set;

An undersampling module, used to undersample and obtain an undersampled set from the majority class sample set;

A first iterative training module, configured to perform first iterative training on a preset classification model based on a training set composed of the minority class sample set and the undersampled set, to obtain a classification model that satisfies the first preset condition;

A detection module configured to detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

An oversampling module, configured to oversample the minority class sample set based on the classification model that satisfies the first preset condition if the second preset condition is not met, and add the oversampled data samples to the training set;

The second iterative training module is configured to perform second iterative training on the classification model that satisfies the first preset condition based on the updated training set to obtain a data classification model that satisfies the second preset condition.

A data classification device including:

The data acquisition module to be classified is used to obtain the data to be classified;

The training device for the above-mentioned data classification model; and,

A classification module, configured to classify the data to be classified using a classification model that meets the preset training stop condition.

A computer device includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, they cause the processor to perform the following steps:

Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set;

Obtain an undersampled set from undersampling the majority class sample set;

Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition;

Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set;

Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition.

A storage medium storing computer-readable instructions that, when executed by one or more processors, cause one or more processors to perform the following steps:

Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set;

Obtain an undersampled set from undersampling the majority class sample set;

Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition;

Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set;

Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition.

The training method, device, computer equipment and storage medium of the above-mentioned data classification model divides multiple historical data samples acquired in advance into a minority class sample set and a majority class sample set, and obtains undersampling from the majority class sample set. set, perform the first iterative training on the preset classification model based on the training set composed of the minority class sample set and the under-sampled set, obtain a classification model that satisfies the first preset condition, and detect that the preset classification model satisfies the first preset condition. Assume whether the conditional classification model satisfies the second preset condition. If it does not satisfy the second preset condition, then oversample the minority class sample set based on the classification model that satisfies the first preset condition, and oversample to obtain The data samples are added to the training set, and the second iterative training is performed on the classification model that satisfies the first preset condition based on the updated training set to obtain the data classification model that satisfies the second preset condition; because when training the classification model Under-sampled data and over-sampled data are used. Therefore, the data used to train the classification model is well balanced, the training effect of the classification model is good, and the trained classification model has high classification accuracy, which overcomes the existing problems. In technology, due to the imbalance of training data used in training classification models, the training effect is poor and the classification accuracy of the trained classification model is low.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the drawings needed to be used in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. , for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative labor.

Figure 1 is an application environment diagram of the training method of the data classification model provided in one embodiment;

Figure 2 is a flow chart of a training method for a data classification model in one embodiment;

Figure 3 is a flow chart of a specific example of a training method for a data classification model;

Figure 4 is a structural block diagram of a training device for a data classification model provided in one embodiment;

Figure 5 is a block diagram of the internal structure of a computer device in one embodiment.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

It will be understood that the terms "first", "second", "third", etc. are only used to differentiate descriptions and cannot be understood as indicating or implying relative importance. It should also be understood that, although the terms "first," "second," "third," etc. are used in the text to describe various elements in some embodiments of the application, these elements should not be limited by these terms. . These terms are only used to distinguish between various elements.

Referring to Figure 1, the training method of the data classification model provided by the embodiment of the present application can be applied in the application environment of Figure 1, where the client can communicate with the server through the network. The server can divide multiple historical data samples obtained from the client into a minority class sample set and a majority class sample set, and obtain an under-sampled set from the majority class sample set, which is composed of a minority class sample set and an under-sampled set. The training set performs the first iterative training on the preset classification model to obtain a classification model that satisfies the first preset condition, and then detects whether the classification model that satisfies the first preset condition satisfies the second preset condition. If it does not meet the The second preset condition is to oversample the minority class sample set based on the classification model that satisfies the first preset condition, add the oversampled data samples to the training set, and based on the updated training set The classification model that satisfies the first preset condition performs a second iterative training to obtain a data classification model that satisfies the second preset condition. Clients can be, but are not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers.

Oversampling and undersampling are two common methods for dealing with unbalanced data. When training a classification model, the oversampling method repeats a very small proportion of minority class data samples many times to increase the number of data samples of this type, while the undersampling method randomly samples a large proportion of majority class data samples to reduce the number of data samples of this type. Number of data samples. Both methods can adjust the number of data samples to balance different categories of data. However, the inventor found that the traditional oversampling method randomly selects a number of minority data samples from the data set to copy and add them to the data set, which can easily cause the classification model to overfit these data samples, which is not conducive to the generalization of the classification model; the traditional oversampling method The undersampling method randomly discards some majority class data samples. These discarded data samples may contain important information. If the classification model loses this information, it cannot accurately identify the category.

Referring to Figure 2, in one embodiment, a training method for a data classification model is proposed, which may include steps S10 to S60:

S10. Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set.

In some embodiments, the plurality of pre-acquired historical data samples include two types of data samples; step S10 may include:

Count the numbers of the two types of data samples in the plurality of historical data samples respectively;

The numbers of the two data samples are compared, a data sample with a smaller number is used to form the minority class sample set, and a data sample with a larger number is used to form the majority class sample set.

For example, the plurality of data samples may include positive data samples and negative data samples, use the first label to mark each data sample belonging to the positive data sample, and use the second label to mark each data belonging to the negative data sample. Samples are labeled. By counting the number of the first label and the second label, the minority class data sample and the majority class data sample can be determined. For example, you can set the first tag to 0 and the second tag to 1. Assume that the number of label 0 is a, the number of label 1 is b, and a is less than b, then the above positive data sample is the minority class data sample, and the negative data sample is the majority class data sample.

Take the telephone customer service scenario as an example. There are very few complaint calls and many consultation calls. The number of calls between the two types differs by hundreds or even a thousand times. Multiple pre-obtained historical telephone customer service data samples are divided into a minority sample set and Majority class sample set, in which the minority class sample set is a collection of complaint phone data samples, and the majority class sample set is a collection of consultation phone data samples. Label 0 can be used to mark complaint phone data samples, and label 1 can be used to mark consultation phone data samples. By counting the number of tags 0 and 1, the number of complaint phone data samples and the number of consultation phone data samples can be determined.

S20. Obtain an under-sampled set from the majority class sample set by under-sampling.

In some implementations, step S20 may include:

A first number of majority class data samples are randomly undersampled from the majority class sample set to form an undersampled set; wherein the absolute value of the difference between the first number and the number of data samples in the minority class sample set is less than Preset threshold.

Referring to Figure 3, in a specific example, assume that the majority class sample set is N, the minority class sample set is P, the undersampling set is N ₀ , the preset undersampling iteration number threshold is m _under , and the preset oversampling iteration The number of times threshold is m _over .

In this specific example, obtaining an undersampled set from the majority class sample set may include:

A first number of majority class data samples are randomly undersampled from N to form a set N ₀ , where the absolute value of the difference between the first number and the number of data samples in P is less than a preset threshold.

且|P|≈|N ₀|。 Randomly sample multiple majority class data samples from N that are similar to the number of samples in P to form a set N ₀ , where,

And |P|≈|N ₀ |.

S30. Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set, to obtain a classification model that satisfies the first preset condition.

In some embodiments, the preset classification model may adopt an existing technology classification model. The first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iterative training in the first iterative training includes:

Using the training set composed of the minority class sample set and the undersampled set to train the current classification model;

Determine whether this training reaches the first preset training times threshold;

If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set;

Determine whether the classification prediction result reaches the first preset accuracy threshold;

If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training.

In some embodiments, using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes:

Utilize the classification model after this training to predict the probability value that each remaining data sample in the majority class sample set belongs to the minority class sample set and the probability value that belongs to the majority class sample set;

The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set.

In the aforementioned specific example, performing first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampling set to obtain a classification model that satisfies the first preset condition may include :

Establish a misclassified sample set E _N ; where the initial misclassified sample set E _N is an empty set;

Use P and N ₀ to train the preset classification model to obtain the trained classification model;

The trained classification model is used to predict the probability distribution of each data sample in different categories in the set NN ₀ , and all data samples whose probability value on the minority data sample category is greater than the preset probability threshold t _N are added to the misclassification Sample set E _N ;

则停止欠采样；否则，合并E _N和N ₀，利用合并得到的集合更新N ₀；其中，合并E _N和N ₀得到N ₀∪E _N，然后利用N ₀∪E _N更新N ₀，即N ₀＝N ₀∪E _N； If the sample set is misclassified

Then stop undersampling; otherwise, merge E _N and N ₀ , and use the merged set to update N ₀ ; where, merge E _N and N ₀ to get N ₀ ∪ E _N , and then use N ₀ ∪ E _N to update N ₀ , that is N ₀ =N ₀ ∪E _N ;

Determine whether the current number of under-sampling reaches the preset under-sampling iteration threshold m _under ; if it does not reach m _under , repeat the above training steps and continue training until the current number of under-sampling reaches m _under and stop training.

In this embodiment, majority class data samples with a similar number to the minority class data samples are randomly undersampled to form a class-balanced training set, and the training set is used to train a preset classification model, and then classification model predictions are gradually added to the training set. Wrong data samples, majority class data samples that are difficult to classify are added to the training set. Therefore, this undersampling method is biased towards retaining majority class data samples that are difficult to classify. These difficult-to-classify data samples often contain important category information. Retaining these difficult-to-classify data samples will help the classification model correctly predict most types of data samples.

S40. Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition.

In some embodiments, the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; step S40 includes:

Use a classification model that satisfies the first preset condition to perform classification prediction on the minority class sample set to obtain a classification prediction result;

Compare the obtained classification prediction result with the second preset accuracy threshold, and determine whether the classification prediction result reaches the second preset accuracy threshold;

If the second preset accuracy threshold is reached, it is determined whether the number of training times reaches the second preset training number threshold.

S50. If the second preset condition is not met, oversample the minority class sample set based on the classification model that satisfies the first preset condition, and add the oversampled data samples to the training set.

In some embodiments, oversampling the minority class sample set based on the classification model that satisfies the first preset condition includes: using the classification model that satisfies the first preset condition to oversample the minority class sample set. Classification prediction: use the data samples with incorrect classification predictions as oversampled data samples based on the classification prediction results.

S60: Perform a second iterative training on the classification model that satisfies the first preset condition based on the updated training set to obtain a data classification model that satisfies the second preset condition.

In some embodiments, each iterative training in the second iterative training includes:

Use the updated training set to train the current classification model;

Determine whether this training reaches the second preset training times threshold;

If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set;

Determine whether the classification prediction result reaches the second preset accuracy threshold;

If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training.

The second preset accuracy threshold can be, for example, 100%, or other accuracy values, which can be set according to actual needs.

In some embodiments, determining whether the classification prediction result reaches a second preset accuracy threshold includes:

Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold.

In the aforementioned example, performing second iterative training on the classification model that satisfies the first preset condition based on the updated training set to obtain a data classification model that satisfies the second preset condition may include:

Establish a minority class sample set P ₀ and initialize P ₀ with P, that is, P ₀ =P;

Establish a misclassified sample set _EP ; where the initial misclassified sample set _EP is an empty set;

The classification model trained by P ₀ and N ₀ is used to predict each data sample in the set P. All data samples with probability values greater than the threshold t _P in the majority category of data samples are added to the misclassified sample set _EP ;

则停止过采样；否则，将E _P中的数据样本加入P ₀； like

Then stop oversampling; otherwise, add the data samples in _EP to P ₀ ;

Determine whether the current number of oversampling reaches the preset threshold m _over for the number of oversampling iterations; if the current number of oversampling does not reach m _over , repeat the above steps until the current number of oversampling reaches m _over and stop.

In this embodiment, a classification model that satisfies the first preset condition is used to predict all minority class data samples, and the data samples with prediction errors are added to the training set repeatedly multiple times, and then the updated training set is used to continue training the classification model, and continue Predict all minority class data samples, and iterate in this manner until all minority class data samples are predicted correctly. Therefore, unlike the random oversampling in the prior art, the oversampling in this embodiment is biased towards enhancing the minority class data samples that are difficult to classify. It is a kind of biased oversampling, which can ensure that the difficulty of classification is enhanced, so as to improve the classification model. The training effect is to obtain a classification model with higher classification accuracy.

In the method of this embodiment, since under-sampled data and over-sampled data are used to train the classification model, the data used to train the classification model is well balanced, and the training effect of the classification model is good. The classification model obtained by training has high classification accuracy, which overcomes the problems in the prior art that the training effect is poor and the classification accuracy of the trained classification model is low due to the imbalance of training data used in training the classification model.

In one embodiment, a data classification method is proposed, including:

S00. Obtain the data to be classified.

Taking the telephone customer service scenario as an example, the data to be classified can be the telephone data received by the customer service, and these telephone data need to be classified into complaint calls and consultation calls.

The steps of the training method of the data classification model in any of the above embodiments; and,

S70: Use the data classification model that satisfies the second preset condition to classify the data to be classified.

Taking the telephone customer service scenario as an example, the data to be classified is input into the data classification model that meets the second preset condition for processing, and the classification result is obtained.

Referring to Figure 4, in one embodiment, a training device for a data classification model is proposed, including:

A division module used to divide multiple pre-obtained historical data samples into a minority class sample set and a majority class sample set;

An undersampling module, used to undersample and obtain an undersampled set from the majority class sample set;

A first iterative training module, configured to perform first iterative training on a preset classification model based on a training set composed of the minority class sample set and the undersampled set, to obtain a classification model that satisfies the first preset condition;

A detection module configured to detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

An oversampling module, configured to oversample the minority class sample set based on the classification model that satisfies the first preset condition if the second preset condition is not met, and add the oversampled data samples to the training set;

The second iterative training module is configured to perform second iterative training on the classification model that satisfies the first preset condition based on the updated training set to obtain a data classification model that satisfies the second preset condition.

In some embodiments, the plurality of pre-acquired historical data samples include two types of data samples; the dividing module is further specifically used to:

Count the numbers of the two types of data samples in the plurality of historical data samples respectively;

The numbers of the two data samples are compared, the smaller number of data samples is used to form the minority class sample set, and the larger number of data samples is used to form the majority class sample set.

In some embodiments, the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iterative training in the first iterative training includes:

Using the training set composed of the minority class sample set and the undersampled set to train the current classification model;

Determine whether this training reaches the first preset training times threshold;

If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set;

Determine whether the classification prediction result reaches the first preset accuracy threshold;

If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training.

In some embodiments, using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes:

Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set;

The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set.

In some embodiments, the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each iterative training in the second iterative training includes:

Use the updated training set to train the current classification model;

Determine whether this training reaches the second preset training times threshold;

If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set;

Determine whether the classification prediction result reaches the second preset accuracy threshold;

If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training.

In some embodiments, determining whether the classification prediction result reaches a second preset accuracy threshold includes:

Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold.

In some implementations, the undersampling module is specifically used to:

A first number of majority class data samples are randomly undersampled from the majority class sample set to form an undersampled set; wherein the absolute value of the difference between the first number and the number of data samples in the minority class sample set is less than Preset threshold.

In one embodiment, a data classification device is provided, including:

The data acquisition module to be classified is used to obtain the data to be classified;

The training device of the data classification model according to any of the above embodiments; and,

A classification module, configured to classify the data to be classified using a classification model that meets the preset training stop condition.

As shown in Figure 5, in one embodiment, a computer device is proposed. The computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor implements the following steps when executing the computer program:

Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set;

Obtain an undersampled set from undersampling the majority class sample set;

Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition;

Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set;

Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition.

In some embodiments, the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each time in the first iterative training executed by the processor Iterative training includes:

Using the training set composed of the minority class sample set and the undersampled set to train the current classification model;

Determine whether this training reaches the first preset training times threshold;

If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set;

Determine whether the classification prediction result reaches the first preset accuracy threshold;

If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training.

In one embodiment, the step performed by the processor to use the trained classification model to perform classification prediction on the remaining data samples in the majority class sample set includes:

Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set;

The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set.

In some embodiments, the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each time in the second iterative training performed by the processor Iterative training includes:

Use the updated training set to train the current classification model;

Determine whether this training reaches the second preset training times threshold;

If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set;

Determine whether the classification prediction result reaches the second preset accuracy threshold;

If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training.

In one embodiment, the step performed by the processor to determine whether the classification prediction result reaches a second preset accuracy threshold includes:

Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold.

In one embodiment, a computer device is proposed. The computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program. Perform the following steps during the program:

Get the data to be classified;

The steps of the training method of the data classification model in any of the above embodiments; and,

The data to be classified is classified using the data classification model that satisfies the second preset condition.

In one embodiment, a storage medium storing computer-readable instructions is proposed. When executed by one or more processors, the computer-readable instructions cause one or more processors to perform the following steps:

Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set;

Obtain an undersampled set from undersampling the majority class sample set;

Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition;

Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition;

If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set;

Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition.

In some embodiments, the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each time in the first iterative training executed by the processor Iterative training includes:

Using the training set composed of the minority class sample set and the undersampled set to train the current classification model;

Determine whether this training reaches the first preset training times threshold;

If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set;

Determine whether the classification prediction result reaches the first preset accuracy threshold;

If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training.

In one embodiment, the step performed by the processor to use the trained classification model to perform classification prediction on the remaining data samples in the majority class sample set includes:

Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set;

The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set.

In some embodiments, the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each time in the second iterative training performed by the processor Iterative training includes:

Use the updated training set to train the current classification model;

Determine whether this training reaches the second preset training times threshold;

If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set;

Determine whether the classification prediction result reaches the second preset accuracy threshold;

If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training.

In one embodiment, the step performed by the processor to determine whether the classification prediction result reaches a second preset accuracy threshold includes:

Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold.

In one embodiment, a storage medium storing computer-readable instructions is proposed. When executed by one or more processors, the computer-readable instructions cause one or more processors to perform the following steps:

Get the data to be classified;

The steps of the training method of the data classification model in any of the above embodiments; and,

The data to be classified is classified using the data classification model that satisfies the second preset condition.

The computer-readable storage medium may be non-volatile or volatile.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. The program can be stored in a computer-readable storage medium. When executed, the process may include the processes of the above method embodiments. Among them, the aforementioned storage media can be non-volatile storage media such as magnetic disks, optical disks, read-only memory (Read-Only Memory, ROM), or random access memory (Random Access Memory, RAM), etc.

The technical features of the above-described embodiments can be combined in any way. To simplify the description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, All should be considered to be within the scope of this manual.

The above-described embodiments only express several implementation modes of the present application, and their descriptions are relatively specific and detailed, but should not be construed as limiting the patent scope of the present application. It should be noted that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all fall within the protection scope of the present application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims (20)

A method for training a data classification model, which includes: Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set; Obtain an undersampled set from undersampling the majority class sample set; Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition; Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition; If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set; Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition. The method of claim 1, wherein the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iterative training in the first iterative training includes : Using the training set composed of the minority class sample set and the undersampled set to train the current classification model; Determine whether this training reaches the first preset training times threshold; If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set; Determine whether the classification prediction result reaches the first preset accuracy threshold; If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training. The method of claim 2, wherein using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes: Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set; The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set. The method of claim 1, wherein the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each iterative training in the second iterative training includes : Use the updated training set to train the current classification model; Determine whether this training reaches the second preset training times threshold; If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set; Determine whether the classification prediction result reaches the second preset accuracy threshold; If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training. The method of claim 4, wherein determining whether the classification prediction result reaches a second preset accuracy threshold includes: Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold. A data classification method, including: Get the data to be classified; The steps of the method according to any one of claims 1-5; and, The data to be classified is classified using the data classification model that satisfies the second preset condition. A training device for a data classification model, which includes: A division module used to divide multiple pre-obtained historical data samples into a minority class sample set and a majority class sample set; An undersampling module, used to undersample and obtain an undersampled set from the majority class sample set; A first iterative training module, configured to perform first iterative training on a preset classification model based on a training set composed of the minority class sample set and the undersampled set, to obtain a classification model that satisfies the first preset condition; A detection module configured to detect whether the classification model that satisfies the first preset condition satisfies the second preset condition; An oversampling module, configured to oversample the minority class sample set based on the classification model that satisfies the first preset condition if the second preset condition is not met, and add the oversampled data samples to the training set; The second iterative training module is configured to perform second iterative training on the classification model that satisfies the first preset condition based on the updated training set to obtain a data classification model that satisfies the second preset condition. The training device of claim 7, wherein the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iterative training in the first iterative training include: Using the training set composed of the minority class sample set and the undersampled set to train the current classification model; Determine whether this training reaches the first preset training times threshold; If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set; Determine whether the classification prediction result reaches the first preset accuracy threshold; If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training. The training device according to claim 8, wherein using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes: Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set; The data samples with classification prediction errors are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set. A data classification device, which includes: The data acquisition module to be classified is used to obtain the data to be classified; The training device according to any one of claims 7-9; and, A classification module, configured to classify the data to be classified using a classification model that meets the preset training stop condition. A computer device includes a memory and a processor. Computer-readable instructions are stored in the memory. When the computer-readable instructions are executed by the processor, they cause the processor to perform the following steps: Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set; Obtain an undersampled set from undersampling the majority class sample set; Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition; Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition; If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set; Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition. The computer device of claim 11, wherein the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iteration training in the first iterative training include: Using the training set composed of the minority class sample set and the undersampled set to train the current classification model; Determine whether this training reaches the first preset training times threshold; If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set; Determine whether the classification prediction result reaches the first preset accuracy threshold; If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training. The computer device according to claim 12, wherein using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes: Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set; The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set. The computer device of claim 11, wherein the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each iteration training in the second iterative training include: Use the updated training set to train the current classification model; Determine whether this training reaches the second preset training times threshold; If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set; Determine whether the classification prediction result reaches the second preset accuracy threshold; If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training. The computer device of claim 14, wherein the determining whether the classification prediction result reaches a second preset accuracy threshold includes: Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold. A storage medium storing computer-readable instructions that, when executed by one or more processors, cause one or more processors to perform the following steps: Divide multiple historical data samples obtained in advance into a minority class sample set and a majority class sample set; Obtain an undersampled set from undersampling the majority class sample set; Perform first iterative training on the preset classification model based on the training set composed of the minority class sample set and the undersampled set to obtain a classification model that satisfies the first preset condition; Detect whether the classification model that satisfies the first preset condition satisfies the second preset condition; If the second preset condition is not met, oversampling the minority class sample set based on the classification model that satisfies the first preset condition, and adding the oversampled data samples to the training set; Based on the updated training set, a second iterative training is performed on the classification model that satisfies the first preset condition to obtain a data classification model that satisfies the second preset condition. The storage medium of claim 16, wherein the first preset condition is reaching a first preset training times threshold or reaching a first preset accuracy threshold; each iteration training in the first iterative training include: Using the training set composed of the minority class sample set and the undersampled set to train the current classification model; Determine whether this training reaches the first preset training times threshold; If the first preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set; Determine whether the classification prediction result reaches the first preset accuracy threshold; If the first preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the under-sampling set to obtain an updated under-sampling set; the updated under-sampling set is used for the first The next iteration of training in iterative training. The storage medium of claim 17, wherein using the classification model after this training to perform classification prediction on the remaining data samples in the majority class sample set includes: Using the classification model after this training to predict the probability value of each remaining data sample in the majority class sample set belonging to the minority class sample set and the probability value of belonging to the majority class sample set; The data samples with wrong classification predictions are data samples whose probability value belongs to the minority class sample set is greater than the probability value belonging to the majority class sample set. The storage medium of claim 16, wherein the second preset condition is reaching a second preset training times threshold or reaching a second preset accuracy threshold; each iteration training in the second iterative training include: Use the updated training set to train the current classification model; Determine whether this training reaches the second preset training times threshold; If the second preset training times threshold is not reached, use the classification model after this training to perform classification prediction on the minority class sample set; Determine whether the classification prediction result reaches the second preset accuracy threshold; If the second preset accuracy threshold is not reached, the data samples with incorrect classification predictions are added to the minority class sample set to obtain an updated minority class sample set; the updated minority class sample set is used as the The updated training set for the next iteration of training in the second iteration of training. The storage medium of claim 19, wherein the determining whether the classification prediction result reaches a second preset accuracy threshold includes: Based on the number of misclassified minority class data samples in the classification prediction result, it is determined whether the classification prediction result reaches a second preset accuracy threshold.

Images